May 21, 2026

Google Antigravity 2.0 review 2026: Browser Subagent, $20/month Pro, and the quota problem

By AICoderScope Team · 11 min read

reviewgoogleantigravityai-idepricingcomparisoncursorwindsurf

Two days after Google I/O 2026, Antigravity 2.0 is the new thing developers are Googling. The Browser Subagent — a Chromium instance that clicks through your web app as it builds it — is the most novel feature shipping in any AI coding tool right now. But four quota cuts in four months from December 2025 to March 2026 left Pro subscribers locked out for days at a time, and the trust damage from that is still fresh.

At $20/month, Antigravity Pro competes directly with Cursor Pro and Windsurf Pro. Here’s the honest assessment after researching the full 2.0 announcement and six months of real-world usage reports.

What Antigravity actually is

Google launched the original Antigravity on November 18, 2025, as a free preview. The pitch was a VS Code fork with agents as the primary interface — not autocomplete-first like Copilot or Tab-first like old Cursor, but autonomous-agent-first from day one.

Antigravity 2.0, announced May 19, 2026 at Google I/O, turns that single IDE into a platform: a rebuilt desktop app, a Go-based CLI, a developer SDK, and a Managed Agents billing tier inside the Gemini API. The model powering it is Gemini 3.5 Flash — Google’s new fast frontier model, launched the same day — running at 289 tokens per second in a version optimized specifically for Antigravity.

The key thing to understand before evaluating pricing: Antigravity is a full IDE replacement, not a plugin for your existing editor. You run Antigravity instead of VS Code, not alongside it. This is the same approach as Cursor, Windsurf, and Kiro — a VS Code fork that gives up Microsoft’s extension marketplace for tighter AI integration. If you’re on JetBrains (IntelliJ, PyCharm, WebStorm), Antigravity has nothing for you.

Pricing in May 2026

At Google I/O 2026, Google restructured the paid tiers and cut the top price:

Plan	Monthly cost	Usage limit vs Pro
Free	$0	Rate-limited (weekly caps)
AI Pro	$20/month	Baseline
AI Ultra — 5x	$100/month	5× Pro limits (new at I/O)
AI Ultra — 20x	$200/month	20× Pro limits (reduced from $249.99)
Managed Agents	Pay-per-run	$25 for 2,500 credits

The $100 Ultra tier is new. The $200 tier is a price cut — Google dropped it from $249.99 at the same event they were trying to rebuild user trust. Whether that’s enough depends heavily on whether the quota history repeats.

One important pricing nuance: Claude models cost extra. Antigravity ships with model switching (Claude Sonnet 4.6, Claude Opus 4.6, GPT-OSS 120B alongside the Gemini lineup), but anything outside Gemini routes through your own API key. Use Opus 4.6 in Antigravity and you’re paying Anthropic’s standard API rates on top of your Antigravity subscription.

What’s new in 2.0

The headline feature everyone is talking about is the Browser Subagent. Antigravity spins up a real Chrome instance, navigates to your app’s local dev server, and actually uses it: clicks buttons, fills forms, takes screenshots, and loops back to fix what breaks. An AI coding tool that runs your code and tells you what it saw is different in kind from one that just generates diffs.

The April 2026 update that preceded 2.0 introduced the Unified Permission System — three tiers (Allow, Ask, Deny) that let you decide what the agent handles autonomously versus what requires your sign-off. Allow safe repetitive actions like Git commits, require confirmation before touching environment files, permanently block specific directories. This was the fix that made Antigravity genuinely usable as a daily driver.

The 2.0 launch adds:

Go-based CLI replacing Gemini CLI. Faster, SSH-compatible, preferences sync with the desktop app bidirectionally.
Mission Control dashboard in the desktop app — a single view for orchestrating multiple parallel agents. Start agent A on the API layer while agent B writes unit tests.
Dynamic subagents — agents spawning specialized sub-agents mid-task for parallel execution.
Scheduled background tasks — cron-style agent runs you configure once and forget.
Voice commands — native voice interface, no third-party integration required.
SDK — host custom agents on your own infrastructure, deploy as Managed Agents inside the Gemini API.
Firebase and Android Studio integrations — first-party connectors that are genuinely tighter than what Cursor offers for Google Cloud workloads.

The Gemini CLI → Antigravity CLI migration is mandatory. Individual users (Google AI Pro, free tier) have until June 18, 2026 to switch; Gemini Code Assist Standard/Enterprise license holders keep Gemini CLI indefinitely.

The quota problem

This is the section most review articles skip. Don’t.

Between December 2025 and March 2026, Google cut Antigravity’s usage quotas four times without advance notice. Pro subscribers ($20/month) went from effectively unlimited Gemini model access to weekly rate caps that hit mid-workday. One developer documented that before January 2026 they used over 300 million input tokens per week on Pro; after the March cuts they hit their weekly cap at under 9 million input tokens.

The resulting backlash filled Google’s AI forum with complaints. The Register covered it in March. Pro users reported 5–7 day lockouts after exceeding weekly caps — not hourly rate limits that reset quickly, but multi-day lockouts that rendered the tool unusable for extended periods.

Google’s response was to add the $100 Ultra tier (launched at I/O 2026) and cut the $200 tier price. These are structural changes to the pricing model, not a commitment to stable limits. There is still no published SLA on what Pro subscribers can expect to consume monthly.

This matters for any developer evaluating Antigravity as a primary tool. Cursor and Windsurf have had pricing changes, but neither has had an extended lockout crisis of this scale. Building your workflow around a tool, then being locked out for a week at $20/month, is not a recoverable experience.

Limitations worth knowing before switching

No JetBrains support. Antigravity is a VS Code fork. The architecture means zero plugin integrations for IntelliJ, PyCharm, WebStorm, or any other JetBrains IDE. If your team runs JetBrains, Antigravity is a non-starter.

Open VSX only. Like Kiro, Antigravity can’t access the official VS Code Marketplace due to licensing. The C# Dev Kit, the official GitHub sign-in extension, and other Microsoft-licensed extensions don’t work. You’re limited to the Open VSX Registry. For most languages this is a minor inconvenience; for C# teams it’s a dealbreaker.

No enterprise security documentation. As of May 2026, Antigravity has no published SOC 2 certification, no GDPR data-processing addendum, and no formal organizational governance controls. Cursor, Copilot, and Tabnine all have these. If your organization has compliance requirements, Antigravity isn’t ready.

Claude models require your own Anthropic key. The multi-model support is real — you can run Claude Opus 4.6 inside Antigravity — but you’re paying Anthropic’s API rates in addition to your Google subscription. Using Opus 4.6 via your own key at standard API pricing for a full coding session can run $5–15 per hour depending on context. The headline “access Claude inside Antigravity” buries that cost.

How Antigravity stacks up

	Antigravity 2.0	Cursor Pro	Windsurf Pro	Claude Code Pro
Price	$20/mo	$20/mo	$20/mo	$20/mo
IDE type	VS Code fork	VS Code fork	VS Code fork	CLI + all IDEs
JetBrains	❌	❌	❌	✅ (extension)
VS Code Marketplace	❌ (Open VSX)	❌ (Open VSX)	❌ (Open VSX)	N/A
Included model	Gemini 3.5 Flash	Cursor-1	SWE-1.6	Sonnet 4.6
Inline completion speed	289 tok/s	Not published	950 tok/s (paid)	N/A
Browser testing agent	✅	❌	❌	❌
Background scheduled tasks	✅	❌	❌	❌
Parallel agent orchestration	✅ (Mission Control)	Limited	Cascade sessions	/batch (parallel)
BYOK model switching	✅ (Claude/GPT/Gemini)	✅	✅	Anthropic models
Quota stability track record	⚠️ (4 cuts, lockouts)	✓	✓	✓
Enterprise security certs	❌	SOC 2	SOC 2	SOC 2
Firebase/GCP integrations	✅ first-party	Plugin	Plugin	Plugin

Windsurf’s SWE-1.6 model runs at 950 tokens per second on paid plans — more than 3× Antigravity’s Gemini 3.5 Flash. That speed difference is noticeable in autocomplete latency. (See our Windsurf IDE review 2026 for a full assessment of SWE-1.6.) Cursor’s newer Cursor-1 model is tuned for edit prediction in a way Gemini Flash isn’t — full breakdown in our Cursor IDE review. Claude Code’s Opus 4.7 hits 87.6% on SWE-bench Verified, the highest published score for any coding agent; see the Claude Code review for cost analysis.

Where Antigravity genuinely wins: the Browser Subagent is not available anywhere else at this price. If you’re building a web app and want the agent to verify it works end-to-end in a browser, Antigravity is the only option. The scheduled background tasks are also unique — no other $20/month tier lets you queue agent runs on a cron schedule.

Honest take

Antigravity 2.0 is the most technically ambitious AI coding tool that launched this week, and also the one with the shakiest track record to back that ambition. The Browser Subagent is a genuine leap — having an agent that sees what your users see closes a loop that every other tool ignores. Mission Control for parallel agents is real and useful for large autonomous tasks.

The quota problem is equally real. Four undisclosed cuts in four months, Pro users locked out for a week — Google broke the implicit contract with early adopters. The I/O 2026 pricing restructure ($100 Ultra tier, $200 price drop) looks like a belated acknowledgment of the problem, not a solution to it.

Use Antigravity Pro if: you’re building Firebase or Android apps where Google’s first-party integrations matter, you want Browser Subagent-powered end-to-end testing in your agent loop, and you’re willing to accept some quota uncertainty while the platform matures.

Stick with Cursor or Windsurf if: you want a stable $20/month tool you can build your daily workflow around without worrying about mid-week lockouts. Both have better autocomplete latency and longer track records of not changing the rules on paying users.

Use Claude Code if: autonomous multi-step correctness matters more than IDE features and you want the highest benchmark scores. It works from every IDE via its CLI and the Max tiers scale cleanly without undisclosed quota changes.

For developers already in the Google Cloud ecosystem — running Cloud Run, Firebase, BigQuery, Android — Antigravity 2.0 is worth a serious look. For everyone else, it’s the second or third tool to reach for after you’ve stabilized your primary workflow.

Visit the official Antigravity site at antigravity.google to start a free account. The Gemini CLI migration deadline for existing individual users is June 18, 2026.

1V1 STARTER KIT · CURSOR

Skip the week of trial-and-error setting up Cursor.

12 production-tested .cursorrules templates, 3 workflow configs, the cost-control checklist. Everything I wish I had on day one.

Get it for $19 (early bird) →

Sources

Last updated May 21, 2026. Pricing and features change frequently; verify current state before purchasing.

Was this article helpful?